
Alright guys, remember last week when Anthropic CEO Dario Amodei went full "Final Boss" at Davos? He didn't just express concern. He literally compared the US selling AI chips to China to "selling nuclear weapons to North Korea."
Naturally, we had to check the receipts. Is Dario just being a drama king, or are Chinese AI firms actually eating Silicon Valley’s lunch while we’re all distracted by the latest Sora clips?
Spoiler alert: It’s the second one. And the data is actually kind of terrifying.
Check this out:
For one, Chinese open-source models controlled a tiny 1.2% of global usage in late 2024. Fast forward to 2025 and they’ve skyrocketed to nearly 30%.
A massive report from OpenRouter and a16z (who looked at 100 trillion tokens of data) shows that Alibaba’s Qwen and DeepSeek’s V3 are now competing head-to-head with OpenAI.
Even the big dogs are jumping ship. Airbnb CEO Brian Chesky publicly ditched ChatGPT for Alibaba’s Qwen because it’s "very good, fast and cheap."
And here’s the craziest part of the story: China is doing this while the US is actively trying to starve them of hardware. Despite the restrictions on Nvidia’s top-tier H100s, Chinese labs are getting "MacGyver-level" creative.
How? According to reports, they use "Mixture of Experts" techniques to squeeze every drop of power out of weaker, export-grade chips. When DeepSeek-R1 dropped in January, it matched OpenAI’s o1 model on benchmarks for a fraction of the cost, and was so impressive it caused Nvidia to lose the most single-day stock value in US history. Even Marc Andreessen called it one of the most amazing breakthroughs he’s ever seen.
So what’s fueling their growth?
While Google and OpenAI keep their models behind a "paywall of secrecy," China releases everything as open source, prices it dirt cheap, then get massive adoption.
Right now on Hugging Face, four of the top ten trending models are either Chinese or built on Chinese foundations. Zhipu AI just IPO’d in Hong Kong and had to cap subscriptions on their GLM-4.7 coding model because demand was too high. The wild part? Their user base isn't just in China: it’s primarily the US, followed by India, and Japan. So yes, American devs are officially using Chinese AI to do real work.
So... Was Dario Right?
Here's the thing: China's AI surge happened DESPITE chip restrictions. Now imagine what happens if they get unrestricted access to Nvidia's H200 chips (which Trump just approved for sale).
As of now, the tech world is split into two very loud camps:
For Team Cautious (Dario): Advanced AI is a digital nuke, and selling chips to rivals is a national security suicide mission.
For Team Business (Jensen Huang): If we don't sell to them, they'll build their own chips faster making the US risk losing both money and influence.
The Big Picture: Google DeepMind’s Demis Hassabis says China is approximately "six months behind." But looking at the leaderboard, that gap is closing in real-time. Whether Dario’s comparison is over-the-top or dead-on depends on one thing: Your view.
So do you think AI is just a cool tech product, or the foundation of global power for the next century?
If it’s the former, selling chips is just good business. If it’s the latter… then yeah, maybe don’t hand the tools to your biggest geopolitical rival.
Either way, China is doing more with less, moving faster, and giving it away for free.
